search

princeton-computational-imaging / Diffusion-SDF

Official code repository for the paper: “Diffusion-SDF: Conditional Generative Modeling of Signed Distance Functions”
MIT License
240 stars 17 forks source link

Recon object not in the object space with SDF-VAE #29

Open ZJU-SJ opened 7 months ago

ZJU-SJ commented 7 months ago

Hi, I am currently endeavoring to train the joint SDF-VAE to acquire SDF modulations. Successful reconstruction is observed when processing data in the object normalization space (as exemplified by ShapeNet). However, when I rotate and translate the object for training, but also normalize the object in the unit cube space. Despite a seemingly well-performing loss, the model struggles to faithfully reconstruct the object. Presented below are the reconstructed results; could this discrepancy be indicative of a limitation in the model? image

GG-Bonds commented 6 months ago

Is this the result of how many epochs you trained? My training effect in the first stage was not very good.